Bug in Tracing instrumentation?

[QnnOkabayashi]

I'm currently authoring a tracing-forest, a Subscriber that performs diagnostics on trace data. We ran into an issue where isahc generates trace data where the entered duration of a parent span is shorter than the entered duration of its children. I've outlined the problem further in this issue. I think I've narrowed down the culprit to these lines:

isahc/src/handler.rs

Lines 403 to 404 in 79338ab

	let span = tracing::trace_span!(parent: &self.span, "header");
	let _enter = span.enter();

Here, the parent is passed in to the child and the child is entered, but the parent is not, causing the children to accumulate time while the parent idles. Is there a reason for not entering the parent span, or is this a bug? If it is intended, what is the motivation for it?

I would guess that the fix would look something like this, where the parent span is entered before the child:

let span = tracing::trace_span!(parent: &self.span, "header");
let _enter = self.span.enter(); // parent enters before the child
let _enter = span.enter();

[sagebind]

Thanks for filing an issue! I'm not actually sure if this is being done by Isahc intentionally or not but is worth digging into. At first guess I suppose maybe I had assumed that entering a child span would cause the parent to be considered "entered" automatically, but maybe that assumption was incorrect. Either way this probably needs fixed.